Trends in Imaging for Suspected Pulmonary Embolism Across US Health Care Systems, 2004 to 2016

This cohort study assesses the use of advanced imaging tests, including chest computed tomography, computed tomographic pulmonary angiography, and ventilation-perfusion scan, for pulmonary embolism from 2004 to 2016.


Introduction
Venous thromboembolism is a common and potentially fatal disease, with an estimated lifetime prevalence of up to 5%. 1 Approximately 20% of individuals with pulmonary embolism (PE) die before diagnosis or on the first day after their diagnosis. 1,2 Because the signs and symptoms of PE are often nonspecific, advanced imaging is commonly used for the diagnosis. 2 Ventilation-perfusion lung studies (V/Q scans, a nuclear medicine examination) were traditionally the noninvasive imaging test of choice, but in the early 2000s this imaging modality was supplanted by chest computed tomographic pulmonary angiography (CTPA). 3,4 Chest CTPA is faster and more sensitive than V/Q scan; however, the improved sensitivity is at least in part believed to be owing to the detection of inconsequential, subsegmental PE. 4 Since its introduction, CTPA has been embraced by emergency department (ED) and hospital physicians, and rates of use have increased dramatically, resulting in concerns of overuse. 5,6 Previous studies examining trends in the use of computed tomography (CT) (based on survey or claims data) have not distinguished chest CT (ie, all chest CT except for CTPA) from CTPA. In 2001, CT (including CTPA) was used in 2.6% of ED visits for chest pain or shortness of breath, which increased to 12.5% in 2009, with an average growth of 28.1% per year. 7 In an analysis of Medicare beneficiaries with suspected PE 8 from 2002 to 2009, chest CT use increased 5-fold, but positivity rates (yield) decreased from 7.3% in 2002 to 5.9% in 2009. This finding suggests that a smaller percentage of patients have received the potential benefit of CTPA with respect to improved detection, and more patients have experienced potential harms, including exposure to ionizing radiation, intravenous contrast, 9,10 and overdiagnosis. This observation is further supported by an increasing incidence of PE, with a lower case mortality rate but no change in overall PE mortality. [10][11][12] In response to calls to curb unnecessary and wasteful diagnostic testing, there have been growing efforts to create and implement decision rules for PE testing that rely on risk stratification algorithms to reduce its unnecessary use. The Wells criteria (combined with D-dimer testing) and the PE rule-out criteria (PERC rule) have been derived, extensively validated, and implemented into clinical practice to identify low-risk patients for whom CTPA can be safely avoided. 2,4,13-15 These risk stratification-based strategies were broadly disseminated through national educational campaigns, such as Choosing Wisely. Five societies (American College of Emergency Physicians, American College of Chest Physicians, American Thoracic Society, Society of Nuclear and Molecular Imaging, and American College of Radiology) published guidelines between 2012 and 2014 promoting the avoidance of CTPA for patients with low probability of PE and a negative D-dimer test result or who are PERC negative. 16,17 The Centers for Medicare & Medicaid Services has also mandated the implementation of these clinical decision rules for Medicare beneficiaries by the creation of clinical decision support tools embedded in the electronic health record at the point of order entry to guide clinicians through risk stratification. 18,19 The implications of these rules for diagnostic imaging use in actual practice is not known. 18,20,21 Implementation studies 22,23 of clinical decision support and clinician feedback have shown improved clinician adherence to guidelines but no reduction in CTPA use, and it is unclear if the use of advanced imaging tests for PE has diminished over time. Prior studies 7,8 have examined CTPA use but were limited to single departments or institutions or did not distinguish chest CTPA from other chest CT or did not cover periods after Choosing Wisely. We assessed the use of chest CT, CTPA, and V/Q scan within 7 US health care systems (including ED, inpatient, and outpatient settings) from 2004 to 2016 to provide a robust assessment of diagnostic imaging use for PE over time.

Data Sources
In this retrospective cohort study, imaging use was obtained from 2004 to 2016 for individuals enrolled in the following 7 US integrated and mixed-model health care systems: Kaiser Permanente All sites have available electronic health care information stored in a virtual data warehouse, including comprehensive capture of all imaging among enrollees. 25,26 Imaging is captured using clinical and administrative data sources, including imaging done within and outside the health care system. The institutional review boards of all collaborating institutions approved the study, and a waiver of individual informed consent was obtained because of the use of deidentified data. The

JAMA Network Open | Imaging
Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline was followed.

Imaging Use
For each calendar year of the study, individuals who initially enrolled, were continuously enrolled, or died during that year were included. Imaging examinations were coded using a combination of Current Procedural Terminology, 27 International Classification of Diseases, Ninth Revision, 28 International Statistical Classification of Diseases, Tenth Revision, 29 and Health Care Financing Administration Common Procedural Coding System 30 billing codes. Examinations were included regardless of the physician specialty billing for the study. We updated a previously used map of billing codes to anatomic area and imaging modality to identify chest CT, chest CTPA, and V/Q nuclear medicine examinations. 31 Within these modalities, chest CTPA for PE was distinguished from chest non-CTPA scans (chest CT) using specific billing codes for PE. There are a total of 117 individual codes (83 CT chest, 9 CTPA, and 25 V/Q scan codes) used to identify chest CT, chest CTPA, and V/Q nuclear medicine examinations. Both professional and technical billing claims were used to capture use; however, to avoid overcounting, only a single imaging examination per person per day was included.

Statistical Analysis
Analyses were conducted between June 11, 2019, and March 18, 2020, and we used all available data from each health care system. Analyses were stratified by imaging test and age group (18-64 years and Ն65 years) and sex (female vs male). We also accounted for follow-up time for individuals based on enrollment for that calendar year. Use rates were modeled with overdispersed Poisson regression, including main effects for examination year and contributing health care system. Absolute annual rates per 1000 person-years with 95% CIs were estimated from these overdispersed Poisson regression models. Rates were averaged across health care systems using equal weights. Relative rate comparisons were made within each period group (2007 vs 2004, 2011 vs 2008, and 2016 vs 2012). All analyses were based on study participants with complete imaging data. If individuals were missing imaging data, they were not included in the analytic cohort.
We used joinpoint regression analysis 32,33 to identify years with statistically significant changes in imaging trends over time and to calculate the mean annual percentage change (growth) within each period and age group. The number of examinations per person-year and SEs estimated from the Poisson regression model were added into the Joinpoint software. A permutation test was used to identify the optimal number of change points for each group (imaging test and age), applying a Bonferroni correction to the type I error to correct for multiple testing. A maximum of 2 change points was allowed based on Joinpoint recommendations given the number of years included.
Annual percentage changes and 95% CIs were estimated assuming that the rates change at a constant percentage every year on a log scale. A second approach also used joinpoint regression but using fixed, specified periods (2004-2007, 2008-2011, and 2012-2016) to generate mean annual percentage change (growth) within each period by imaging test and age group, allowing easier reporting in tabular format and comparisons across period age group. Poisson regression analyses were conducted using SAS, version 9.4 (SAS Institute Inc), and joinpoint regression was performed using Joinpoint Regression Program, version 4.7.0.0 (National Cancer Institute). 34

JAMA Network Open | Imaging
Trends in Imaging for Suspected Pulmonary Embolism Across US Health Care Systems

Results
Overall, 3.6 to 4.8 million enrollees were included in each year of the study, for a total of 52 343 517 person-years of follow-up data ( CTPA examinations (16.5% of all of the CTs performed in the chest), and 59 208 V/Q scans. Among the 7 health systems, the rate of chest CT ranged from 29.3 to 50.6 per 1000 person-years, the rate of CTPA ranged from 5.3 to 8.6 per 1000 person-years, and the rate of V/Q scans ranged from 0.8 to 1.8 per 1000 person-years across years. Chest CT scan rates were higher than CTPA examination rates (on average, 4 to 8 times higher across health systems), and both were higher than V/Q scan rates. Averaging across the entire study period, imaging rates for each of the 3 tests were approximately 4 times higher in adults 65 years or older compared with adults aged 18 to 64 years.
Imaging with chest CT and CTPA increased over time, whereas V/Q scan use decreased over time (  The average growth using fixed periods is summarized in Table 2

Discussion
The use of CTPA for suspected PE across 7 US integrated health care systems has continued to increase in recent years. Annual growth in CTPA was highest in the earlier years of the study (ie, from 2004 to 2006), but imaging with CTPA has continued to increase in both adults aged 18 to 64 years and adults 65 years or older. In contrast, nuclear medicine imaging had a consistent decline in both age categories and across all health care systems. However, the growth in CTPA far outpaced the decline in V/Q scanning.
Previous studies 35,36 of diagnostic imaging use in large US health care systems have found that older patients have higher rates of advanced imaging overall compared with younger patients.
Imaging use increased steeply with age for CT, nuclear medicine, and magnetic resonance imaging, but in particular for CT scan. 35,36 In the previous study 36 of imaging use across the same 7 US integrated and mixed-model health care systems, CT imaging rates per person-years were highest in adults 65 years or older across most anatomic regions. Findings in the present study demonstrate a similar pattern for the use of CTPA to screen for PE. The exact reasons for the higher rates of imaging in adults 65 years or older are unclear but are likely multifactorial. It is well known that age is a risk factor for both PE and mortality in those diagnosed as having PE. This risk is reflected in clinical decision rules, such as the PERC rule, in which age 50 years or older is a risk factor that prevents exclusion of PE on clinical grounds. D-dimer testing is used to screen for PE and is more likely to be elevated in older patients than in younger patients. In addition, although cancer risk from radiation is often thought to decline with age, models suggest that cancer risk declines with age until middle age, when cancer risk may then increase in a U-shaped distribution. 37,38 Therefore, radiation-related cancer risk after exposure in middle and older ages may be higher than previously believed. The investigators reported steadily increasing rates of CT use, although their analysis was limited to a specific age range (in most cases, adults 65 years or older). Results in the present study reflect imaging from 2004 through 2016 among adults of all ages at 7 health care systems across the United States, and we were able to differentiate the use of CTPA from the use of chest CT, as well as assess the use of V/Q scan.
Both CT scan and CT angiography are valuable diagnostic tests that in many cases has led to accurate diagnoses and improved patient outcomes. 2 However, 1 in 4 Americans receive CT scans each year, and many of these tests are unnecessary and expose patients to a number of risks, including anxiety and discomfort, ionizing radiation, and incidental findings. 10 Exposure to radiation is believed to increase the lifetime risk of cancer, 10 and the National Cancer Institute has estimated that 2% of cancers are iatrogenic. 9 Incidental findings often lead to additional testing and unnecessary procedures. A study 40 of Medicare beneficiaries found a statistically significant association between the number of CT scans of the abdomen or pelvis and the performance of nephrectomy. There are a number of reasons for the increase in CT use over time. The technology of CT scanning continues to improve, resulting in faster, more accurate studies. Clinicians increasingly rely on CT scans to avoid missing serious conditions and incurring malpractice claims. Payment models have incentivized clinicians and health systems to perform imaging, with few disincentives to request low-yield or inappropriate studies.

JAMA Network Open | Imaging
Trends in Imaging for Suspected Pulmonary Embolism Across US Health Care Systems Several studies have found that well-conducted and validated approaches to reduce CTPA overuse are not having the desired impact. 2, 14,15 As demonstrated in the present study, not only have CTPA imaging rates not declined, but they have also continued to show 3.0% to 4.3% annual growth through 2016. The Wells criteria and D-dimer testing strategies have been modified over the last 2 decades, reformulated from 3 levels of risk (low, moderate, and high) to 2 levels (unlikely vs likely) to simplify decision-making at the bedside. 41 D-dimer cutoffs have been altered to include age adjustment to better identify older patients who are at low risk. 14 The YEARS study algorithm was derived and validated to produce a decision rule for suspected PE, with few items to simplify score calculation. 42 Despite efforts and national initiatives to disseminate and implement these research findings, 2,14,15,42 we found that CTPA use to screen for PE continues to increase. It is not clear why such efforts have not curtailed the growth in CTPA imaging, except to note that contributors to CT use persist, such as fear of missing PE, concerns about malpractice, 43 improvements in technology and CT availability, and financial incentives. Also, it may be that approaches to developing risk stratification tools have been overly conservative, and the goal to create decision tools with high sensitivity may render such tools inefficient in that they may not identify a large proportion of low-risk patients in whom testing can be deferred. The Wells criteria for suspected PE contain a heavily weighted subjective component, which although providing flexibility might be interpreted in an overly cautious manner by clinicians. Finally, studies of the implementation of these clinical decision rules have not been conducted using optimal randomized designs, resulting in an inability to rigorously measure their impact on imaging overuse.
These observations suggest that the process of incorporating imaging tests into clinical practice should be governed by rigorous analysis of the benefits and harms for patients and health systems.
The use of CTPA was rapidly embraced as the preferred first-line diagnostic test for suspected PE soon after reports touted its greater sensitivity than V/Q scan 4 but before other outcomes were assessed, such as the rate of incidental diagnoses and their potential for overtreatment, as well as before the prognosis of subsegmental thrombi was understood. 6,11 Furthermore, the ready availability of CT scanning likely promotes its use as a first-line test over others, such as the V/Q scan. Therefore, the continued use of CTPA as the first-line imaging test for PE may not have been guided by a balanced consideration of all benefits and harms. However, as shown by the results herein, deimplementation of tests once adopted is difficult, even when there is widespread agreement that the tests are overused.

Limitations
This study has several limitations. First, the study included US patients enrolled in a limited number of health care systems, all of which used health maintenance organization models of care either in part or in whole. These health care systems were chosen because they are members of the National Cancer Institute-supported Health Care Systems Research Network and collect data in a common format that are locally stored in virtual data warehouses. These systems are diverse; however, results in the present study are limited because all of the included sites are integrated health care systems that used health maintenance organization models of care either in part or in whole. Patterns of imaging over time among these patients may not represent patterns among individuals covered by fee-for-service plans with different incentives and disincentives. For example, an increase in high deductibles might have diminished the use of expensive tests in fee-for-service plans. Second, the indication for imaging was not available in the data set herein, nor were the results of any risk stratification or D-dimer testing. Therefore, it was not possible to assess whether imaging was appropriate or inappropriate or whether imaging use was associated with improved patient outcomes. Third, we used specific imaging codes to identify CTPA for suspected PE, and some PE studies may have been coded as routine chest CT. Nonetheless, imaging patterns for CTPA paralleled the overall pattern of chest CT, with continued increase in recent years. Therefore, it is unlikely that undercapturing of some examinations performed to screen for PE would have altered the overall conclusions.